From 37a43e245f2e901d46c8cacf948d8909c1d343a5 Mon Sep 17 00:00:00 2001
From: wheremyfoodat <44909372+wheremyfoodat@users.noreply.github.com>
Date: Sun, 25 Aug 2024 19:13:37 +0300
Subject: [PATCH] HW shaders: Accelerate indexed draws

---
 include/renderer_gl/renderer_gl.hpp  |  8 +++++++-
 src/core/PICA/gpu.cpp                |  2 +-
 src/core/renderer_gl/renderer_gl.cpp | 26 +++++++++++++++++---------
 3 files changed, 25 insertions(+), 11 deletions(-)

diff --git a/include/renderer_gl/renderer_gl.hpp b/include/renderer_gl/renderer_gl.hpp
index b643534a..30b17026 100644
--- a/include/renderer_gl/renderer_gl.hpp
+++ b/include/renderer_gl/renderer_gl.hpp
@@ -60,9 +60,15 @@ class RendererGL final : public Renderer {
 	float oldDepthScale = -1.0;
 	float oldDepthOffset = 0.0;
 	bool oldDepthmapEnable = false;
-	// Set by prepareDraw, tells us whether the current draw is using hw-accelerated shader
+	// Set by prepareForDraw, tells us whether the current draw is using hw-accelerated shader
 	bool usingAcceleratedShader = false;
 	bool performIndexedRender = false;
+	bool usingShortIndices = false;
+
+	// Set by prepareForDraw, metadata for indexed renders
+	GLuint minimumIndex = 0;
+	GLuint maximumIndex = 0;
+	void* hwIndexBufferOffset = nullptr;
 
 	// Cached pointer to the current vertex shader when using HW accelerated shaders
 	OpenGL::Shader* generatedVertexShader = nullptr;
diff --git a/src/core/PICA/gpu.cpp b/src/core/PICA/gpu.cpp
index dad24a22..2797e09f 100644
--- a/src/core/PICA/gpu.cpp
+++ b/src/core/PICA/gpu.cpp
@@ -141,7 +141,7 @@ void GPU::drawArrays(bool indexed) {
 		// Total # of vertices to render
 		const u32 vertexCount = regs[PICA::InternalRegs::VertexCountReg];
 
-		// Note: In the hardware shader path the vertices span shouldn't actually be used as the rasterizer will perform its own attribute fetching
+		// Note: In the hardware shader path the vertices span shouldn't actually be used as the renderer will perform its own attribute fetching
 		renderer->drawVertices(primType, std::span(vertices).first(vertexCount));
 	} else {
 		const bool shaderJITEnabled = ShaderJIT::isAvailable() && config.shaderJitEnabled;
diff --git a/src/core/renderer_gl/renderer_gl.cpp b/src/core/renderer_gl/renderer_gl.cpp
index 82248d53..536cb6fa 100644
--- a/src/core/renderer_gl/renderer_gl.cpp
+++ b/src/core/renderer_gl/renderer_gl.cpp
@@ -435,10 +435,8 @@ void RendererGL::drawVertices(PICA::PrimType primType, std::span<const Vertex> v
 	const auto primitiveTopology = primTypes[static_cast<usize>(primType)];
 	gl.disableScissor();
 
-	if (usingAcceleratedShader) {
-		hwVertexBuffer->Bind();
-		gl.bindVAO(hwShaderVAO);
-	} else {
+	// If we're using accelerated shaders, the hw VAO, VBO and EBO objects will have already been bound in prepareForDraw
+	if (!usingAcceleratedShader) {
 		vbo.bind();
 		gl.bindVAO(defaultVAO);
 	}
@@ -509,9 +507,12 @@ void RendererGL::drawVertices(PICA::PrimType primType, std::span<const Vertex> v
 		OpenGL::draw(primitiveTopology, GLsizei(vertices.size()));
 	} else {
 		if (performIndexedRender) {
-			// When doing indexed rendering, bind the IBO and use glDrawRangeElementsBaseVertex to issue the indexed draw
+			// When doing indexed rendering, bind the EBO and use glDrawRangeElementsBaseVertex to issue the indexed draw
 			hwIndexBuffer->Bind();
-			//glDrawRangeElementsBaseVertex();
+			glDrawRangeElementsBaseVertex(
+				primitiveTopology, minimumIndex, maximumIndex, GLsizei(vertices.size()), usingShortIndices ? GL_UNSIGNED_SHORT : GL_UNSIGNED_BYTE,
+				hwIndexBufferOffset, -minimumIndex
+			);
 		} else {
 			// When doing non-indexed rendering, just use glDrawArrays
 			OpenGL::draw(primitiveTopology, GLsizei(vertices.size()));
@@ -1008,7 +1009,10 @@ bool RendererGL::prepareForDraw(ShaderUnit& shaderUnit, PICA::DrawAcceleration*
 
 			// Upload vertex data and index buffer data to our GPU
 			accelerateVertexUpload(shaderUnit, accel);
+
 			performIndexedRender = accel->indexed;
+			minimumIndex = GLsizei(accel->minimumIndex);
+			maximumIndex = GLsizei(accel->maximumIndex);
 		}
 	}
 
@@ -1146,17 +1150,21 @@ void RendererGL::accelerateVertexUpload(ShaderUnit& shaderUnit, PICA::DrawAccele
 
 	// Update index buffer if necessary
 	if (accel->indexed) {
-		const bool shortIndex = accel->useShortIndices;
-		const usize indexBufferSize = usize(vertexCount) * (shortIndex ? sizeof(u16) : sizeof(u8));
+		usingShortIndices = accel->useShortIndices;
+		const usize indexBufferSize = usize(vertexCount) * (usingShortIndices ? sizeof(u16) : sizeof(u8));
 
+		hwIndexBuffer->Bind();
 		auto indexBufferRes = hwIndexBuffer->Map(4, indexBufferSize);
+		hwIndexBufferOffset = reinterpret_cast<void*>(usize(indexBufferRes.buffer_offset));
+
 		std::memcpy(indexBufferRes.pointer, accel->indexBuffer, indexBufferSize);
 		hwIndexBuffer->Unmap(indexBufferSize);
 	}
 
+	hwVertexBuffer->Bind();
 	auto vertexBufferRes = hwVertexBuffer->Map(4, accel->vertexDataSize);
-
 	u8* vertexData = static_cast<u8*>(vertexBufferRes.pointer);
+
 	gl.bindVAO(hwShaderVAO);
 
 	for (int i = 0; i < totalAttribCount; i++) {